[net,tr,Y,​E]=train(n​et,P,T); How Y and E get their dimensions with this syntax?

Hello

I am training a neural network with EEG dataset.

I have Input Matrix P with [m n] = size(P) m = 100 n = 45

and I manually created target T as [m n] = size(T) m = 100 n = 45

after this I am training my network with code below:

 net=newff(P,T,2,{'logsig','purelin'},'traingd');
net.trainParam.show = 30;
net.trainParam.lr = 0.01;
net.trainParam.epochs = 30000;
net.trainParam.goal = 0;
[net,tr,Y,E]=train(net,P,T);

After this stage, I get Y and E as:

[m n] = size(Y) m = 100 n = 27

and [m n]= size(E) m = 100 n = 27

I am not able to understand how it is deciding n=27.

This value is creating trouble ahead for me as I move forward with my code:

%% Evaluate network performance and plot results % evaluate performance: decoding network response

[m,i] = max(T); % target class

[m,j] = max(Y); % predicted class

Ns = length(Y); % number of all samples

k = 0; % number of missclassified samples

if find(i-j), % if there exist missclassified samples

 k = length(find(i-j)); % get a number of missclassified samples

end

fprintf('Correct classified samples: %.1f%% samples\n', 100*(N-k)/N)

it is not able to execute if statement as i and j are not of the same size and I am getting an error

??? Error using ==> minus Matrix dimensions must agree.

Please suggest a working solution. I am using Matlab R2010a so cannot proceed with feedforward net.

[Greg Heath]

Don't know.

However, typically you should use n >> m which is usually interpreted as: the number of samples to be at least factor of 10 to 30 times the dimensionality of the input vectors

[JAGRITI SAINI]

This problem is solved now... I did it with other syntax:

[net]=train(net,P,T);

and calculated y separately.

Y = sim(net,P);

It helped me to get classified samples finally.

[Mohamed Nedal]

Hello, in the LHS "[net,tr,Y,E]", What does "E" stands for? Is that the error of the network?

[Greg Heath]

Yes:

 E = T-Y;

Greg

Answer 1

[Greg Heath]

For pattern recognition use NEWPR which is the special case of NEWFF that is designed for classification and pattern recognition

(NEWFIT is the special case for curve fitting and regression).

For classification of N I-dimensional "I"nput vectors into c distinct classes, the N target vectors should be [0,1] c-dimensional unit vectors.

[ I N ] = size(input)
[ c N ] = size(target)

with sum(target) = ones(1,N)

For examples search the NEWSGROUP and ANSWERS with

 greg NEWPR

also check the similar use of the latest classifier

 greg PATTERNNET

Hope this helps.

Thank you for formally accepting my answer

Greg